diff mbox

[v2] module_to_cil: fix use of uninitialized variables

Message ID 1470260523-2511-1-git-send-email-william.c.roberts@intel.com (mailing list archive)
State Not Applicable
Headers show

Commit Message

Roberts, William C Aug. 3, 2016, 9:42 p.m. UTC
From: William Roberts <william.c.roberts@intel.com>

Correct errors like these reported by gcc:

module_to_cil.c: In function ‘block_to_cil’:
module_to_cil.c:229:20: error: ‘attr_list’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
  struct list_node *curr = (*attr_list)->head;

Usages of attr_list_destroy() were called when list_init()
fails. Just bail early on failure.

stack_init() and stack_destroy() also suffered from the
aforementioned issue.

Signed-off-by: William Roberts <william.c.roberts@intel.com>
---
 libsepol/src/module_to_cil.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

Comments

Steve Lawrence Aug. 3, 2016, 9:51 p.m. UTC | #1
On 08/03/2016 05:42 PM, william.c.roberts@intel.com wrote:
> From: William Roberts <william.c.roberts@intel.com>
> 
> Correct errors like these reported by gcc:
> 
> module_to_cil.c: In function ‘block_to_cil’:
> module_to_cil.c:229:20: error: ‘attr_list’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
>   struct list_node *curr = (*attr_list)->head;
> 
> Usages of attr_list_destroy() were called when list_init()
> fails. Just bail early on failure.
> 
> stack_init() and stack_destroy() also suffered from the
> aforementioned issue.
> 
> Signed-off-by: William Roberts <william.c.roberts@intel.com>
> ---
>  libsepol/src/module_to_cil.c | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/libsepol/src/module_to_cil.c b/libsepol/src/module_to_cil.c
> index b9a4af7..9d0d064 100644
> --- a/libsepol/src/module_to_cil.c
> +++ b/libsepol/src/module_to_cil.c
> @@ -1307,7 +1307,7 @@ static int cond_list_to_cil(int indent, struct policydb *pdb, struct cond_node *
>  
>  	rc = list_init(&attr_list);
>  	if (rc != 0) {
> -		goto exit;
> +		return rc;
>  	}
>  
>  	for (cond = cond_list; cond != NULL; cond = cond->next) {
> @@ -3488,7 +3488,7 @@ static int block_to_cil(struct policydb *pdb, struct avrule_block *block, struct
>  
>  	rc = list_init(&attr_list);
>  	if (rc != 0) {
> -		goto exit;
> +		return rc;
>  	}
>  
>  	rc = typealiases_to_cil(indent, pdb, block, stack);
> @@ -3635,7 +3635,7 @@ static int blocks_to_cil(struct policydb *pdb)
>  
>  	rc = stack_init(&stack);
>  	if (rc != 0) {
> -		goto exit;
> +		return rc;
>  	}
>  
>  	block = pdb->global;
> @@ -3703,7 +3703,7 @@ static int linked_blocks_to_cil(struct policydb *pdb)
>  
>  	rc = stack_init(&stack);
>  	if (rc != 0) {
> -		goto exit;
> +		return rc;
>  	}
>  
>  	block = pdb->global;
> 

I would recommend just initializing the variables to NULL and keeping
the "goto exit"'s. That would maintain the single return point, allows
for extra cleanup code to be run in the future if necessary, and is
consistent with the rest of the module_to_cil code.

- Steve
Roberts, William C Aug. 3, 2016, 9:59 p.m. UTC | #2
> -----Original Message-----
> From: Steve Lawrence [mailto:slawrence@tresys.com]
> Sent: Wednesday, August 3, 2016 2:51 PM
> To: Roberts, William C <william.c.roberts@intel.com>; selinux@tycho.nsa.gov;
> seandroid-list@tycho.nsa.gov; sds@tycho.nsa.gov
> Subject: Re: [PATCH v2] module_to_cil: fix use of uninitialized variables
> 
> On 08/03/2016 05:42 PM, william.c.roberts@intel.com wrote:
> > From: William Roberts <william.c.roberts@intel.com>
> >
> > Correct errors like these reported by gcc:
> >
> > module_to_cil.c: In function ‘block_to_cil’:
> > module_to_cil.c:229:20: error: ‘attr_list’ may be used uninitialized in this
> function [-Werror=maybe-uninitialized]
> >   struct list_node *curr = (*attr_list)->head;
> >
> > Usages of attr_list_destroy() were called when list_init() fails. Just
> > bail early on failure.
> >
> > stack_init() and stack_destroy() also suffered from the aforementioned
> > issue.
> >
> > Signed-off-by: William Roberts <william.c.roberts@intel.com>
> > ---
> >  libsepol/src/module_to_cil.c | 8 ++++----
> >  1 file changed, 4 insertions(+), 4 deletions(-)
> >
> > diff --git a/libsepol/src/module_to_cil.c
> > b/libsepol/src/module_to_cil.c index b9a4af7..9d0d064 100644
> > --- a/libsepol/src/module_to_cil.c
> > +++ b/libsepol/src/module_to_cil.c
> > @@ -1307,7 +1307,7 @@ static int cond_list_to_cil(int indent, struct
> > policydb *pdb, struct cond_node *
> >
> >  	rc = list_init(&attr_list);
> >  	if (rc != 0) {
> > -		goto exit;
> > +		return rc;
> >  	}
> >
> >  	for (cond = cond_list; cond != NULL; cond = cond->next) { @@ -3488,7
> > +3488,7 @@ static int block_to_cil(struct policydb *pdb, struct
> > avrule_block *block, struct
> >
> >  	rc = list_init(&attr_list);
> >  	if (rc != 0) {
> > -		goto exit;
> > +		return rc;
> >  	}
> >
> >  	rc = typealiases_to_cil(indent, pdb, block, stack); @@ -3635,7
> > +3635,7 @@ static int blocks_to_cil(struct policydb *pdb)
> >
> >  	rc = stack_init(&stack);
> >  	if (rc != 0) {
> > -		goto exit;
> > +		return rc;
> >  	}
> >
> >  	block = pdb->global;
> > @@ -3703,7 +3703,7 @@ static int linked_blocks_to_cil(struct policydb
> > *pdb)
> >
> >  	rc = stack_init(&stack);
> >  	if (rc != 0) {
> > -		goto exit;
> > +		return rc;
> >  	}
> >
> >  	block = pdb->global;
> >
> 
> I would recommend just initializing the variables to NULL and keeping the "goto
> exit"'s. That would maintain the single return point, allows for extra cleanup code
> to be run in the future if necessary, and is consistent with the rest of the
> module_to_cil code.

That's fine, but requiring only one exit point isn't consistent with the rest of libsepol,
some do it this way, some don’t. There will be slightly more overhead In the approach
of initializing it first and then jumping on failure, then checking if its NULL in the
destruction routine, then finally returning versus just checking the init routine
and returning.

> 
> - Steve
>
William Roberts Aug. 4, 2016, 9:43 p.m. UTC | #3
<snip>
>
> I would recommend just initializing the variables to NULL and keeping
> the "goto exit"'s. That would maintain the single return point, allows
> for extra cleanup code to be run in the future if necessary, and is
> consistent with the rest of the module_to_cil code.
>

FYI these functions have multiple return paths, so that statement isn't
even true for module_to_cil code:

roles_gather_map()
typealiases_gather_map()
stack_destroy()
is_id_in_scope_with_start()


Functions that always return 0:
semantic_level_to_cil()
cil_print_attr_strs()


Some functions have double sets to -1 on rc, the compiler should be smart
enough to disregard those, but
still, no consistency on when/where rc = -1 is set.

ebitmap_to_names()

I only went through 36% of the file and noticed all these.

On a separate note, Coverity has free scanning for open source projects,
perhaps we should put selinux project in there
and get it scanning. It will likely pick up a lot of stuff.
diff mbox

Patch

diff --git a/libsepol/src/module_to_cil.c b/libsepol/src/module_to_cil.c
index b9a4af7..9d0d064 100644
--- a/libsepol/src/module_to_cil.c
+++ b/libsepol/src/module_to_cil.c
@@ -1307,7 +1307,7 @@  static int cond_list_to_cil(int indent, struct policydb *pdb, struct cond_node *
 
 	rc = list_init(&attr_list);
 	if (rc != 0) {
-		goto exit;
+		return rc;
 	}
 
 	for (cond = cond_list; cond != NULL; cond = cond->next) {
@@ -3488,7 +3488,7 @@  static int block_to_cil(struct policydb *pdb, struct avrule_block *block, struct
 
 	rc = list_init(&attr_list);
 	if (rc != 0) {
-		goto exit;
+		return rc;
 	}
 
 	rc = typealiases_to_cil(indent, pdb, block, stack);
@@ -3635,7 +3635,7 @@  static int blocks_to_cil(struct policydb *pdb)
 
 	rc = stack_init(&stack);
 	if (rc != 0) {
-		goto exit;
+		return rc;
 	}
 
 	block = pdb->global;
@@ -3703,7 +3703,7 @@  static int linked_blocks_to_cil(struct policydb *pdb)
 
 	rc = stack_init(&stack);
 	if (rc != 0) {
-		goto exit;
+		return rc;
 	}
 
 	block = pdb->global;